CN116305285B - Patient information desensitization processing method and system combined with artificial intelligence - Google Patents

Abstract

The invention relates to the technical field of artificial intelligence and information processing, in particular to a patient information desensitization processing method and system combining the artificial intelligence. According to the patient information desensitization processing method and system combining artificial intelligence, the text layout similarity of the target desensitization text and the patient medical record text to be desensitized in the whole layer can be guaranteed, the privacy paragraph information similarity of the target desensitization text and the patient medical record text to be desensitized can also be guaranteed, the matching property between the target desensitization text and the patient medical record text to be desensitized is improved, and therefore the data anonymization/desensitization quality and efficiency of the patient medical record text to be desensitized are improved when the patient medical record text to be desensitized is subjected to data anonymization/desensitization based on the target desensitization text.

Description

Patient information desensitization processing method and system combined with artificial intelligence

Technical Field

The present invention relates to the field of artificial intelligence and information processing technology, and in particular to a patient information desensitization processing method and system combined with artificial intelligence.

Background technique

Information desensitization refers to the deformation of certain sensitive information through desensitization rules to achieve reliable protection of sensitive privacy data. In the case of security data or some sensitive data, the real data is transformed and provided for certain use without violating the system rules. In the field of smart medical care, patient information usually carries some patient privacy, so it is very necessary to desensitize patient information.

Summary of the invention

In order to improve the technical problems existing in the related technologies, the present invention provides a patient information desensitization processing method and system combined with artificial intelligence.

In a first aspect, an embodiment of the present invention provides a patient information desensitization processing method combined with artificial intelligence, which is applied to an AI desensitization processing system, and the method includes: obtaining a medical record text of a patient to be desensitized and a desensitized medical record text of a patient;

Obtain the parsing results of the private paragraphs in the medical record text to be desensitized and the parsing results of the private paragraphs in the desensitized medical record text;

Determine the privacy paragraph commonality score of the medical record text to be desensitized and the desensitized medical record text based on the privacy paragraph parsing results in the medical record text to be desensitized and the privacy paragraph parsing results in the desensitized medical record text of the patient;

Determine the text semantic features of the medical record text to be desensitized and the text semantic features of the desensitized medical record text;

Determine a first semantic commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record based on the text semantic features of the medical record text to be desensitized and the text semantic features of the desensitized text of the patient's medical record;

Determine a text commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record based on the privacy paragraph commonality score of the medical record text to be desensitized and the desensitized text of the patient's medical record and the first semantic commonality score;

Based on the text commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record, a target desensitized text is determined from the desensitized text of the patient's medical record, so as to perform privacy desensitization protection on the medical record text to be desensitized based on the target desensitized text.

In some embodiments, the privacy paragraph parsing result in the medical record text to be desensitized includes a first privacy paragraph text mined from the medical record text to be desensitized, and the privacy paragraph parsing result in the medical record desensitized text includes a second privacy paragraph text mined from the medical record desensitized text; wherein, based on the privacy paragraph parsing result in the medical record text to be desensitized and the privacy paragraph parsing result in the medical record desensitized text, determining the privacy paragraph commonality score of the medical record text to be desensitized and the medical record desensitized text includes:

Determine first statistical data of a target private paragraph in the medical record text to be desensitized based on the first private paragraph text;

Determine second statistical data of the target private paragraph in the desensitized text of the patient's medical record based on the second private paragraph text;

Based on the first statistical data and the second statistical data, determine a sum and a difference of the number of target private paragraphs in the medical record text to be desensitized and the desensitized medical record text;

Determine a commonality score of the number of private paragraphs in the medical record text to be desensitized and the desensitized text of the patient's medical record based on the sum and difference of the number of target private paragraphs in the medical record text to be desensitized and the desensitized text of the patient's medical record;

Based on the commonality score of the number of private paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record, the commonality score of the private paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record is determined.

In some embodiments, obtaining the parsing result of the private paragraphs in the medical record text to be desensitized and the parsing result of the private paragraphs in the desensitized medical record text includes:

Performing privacy paragraph mining processing on the medical record text to be desensitized and the desensitized text of the patient's medical record, respectively, to determine a first privacy paragraph text included in the medical record text to be desensitized and a second privacy paragraph text included in the desensitized text of the patient's medical record;

Based on the first privacy paragraph text and the second privacy paragraph text, determine the privacy paragraph identifier to which each text unit in the medical record text to be desensitized belongs and the privacy paragraph identifier to which each text unit in the medical record desensitized text belongs;

Decomposing the medical record text to be desensitized into X first text sets, and decomposing the medical record desensitized text into X second text sets, wherein the X first text sets correspond to the X second text sets one by one, and X is an integer not less than 1;

Determine the first privacy paragraph identifier of each text unit in each first text set based on the privacy paragraph identifier of each text unit in the medical record text to be desensitized, as the privacy paragraph parsing result in the medical record text to be desensitized;

Based on the privacy paragraph identifiers of each text unit in the desensitized text of the patient's medical record, the second privacy paragraph identifiers of each text unit in each second text set are determined as the privacy paragraph parsing results in the desensitized text of the patient's medical record.

In some embodiments, based on the privacy paragraph parsing results in the medical record text to be desensitized and the privacy paragraph parsing results in the desensitized text of the patient's medical record, determining the privacy paragraph commonality score of the medical record text to be desensitized and the desensitized text of the patient's medical record, includes:

Determine a privacy paragraph identifier commonality score between each first text set and the corresponding second text set based on a first privacy paragraph identifier to which each text unit belongs in each first text set and a second privacy paragraph identifier to which each text unit belongs in each second text set;

Determine the privacy paragraph topic commonality score between the patient medical record text to be desensitized and the patient medical record desensitized text based on the privacy paragraph identification commonality score between each first text set and the corresponding second text set;

Based on the commonality scores of the private paragraph topics of the medical record text to be desensitized and the desensitized text of the patient's medical record, the commonality scores of the private paragraphs of the medical record text to be desensitized and the desensitized text of the patient's medical record are determined.

In some embodiments, the first privacy paragraph identifier includes a first target privacy paragraph identifier, the second privacy paragraph identifier includes a second target privacy paragraph identifier, the first text set includes a first target text set, the second text set includes a second target text set, the first target text set corresponds to the second target text set, a text unit in the first target text set belongs to the first target privacy paragraph identifier, and a text unit in the second target text set belongs to the second target privacy paragraph identifier; wherein determining the privacy paragraph identifier commonality score between each first text set and the corresponding second text set includes:

Determine the number of identical privacy paragraph identifiers between the first target privacy paragraph identifier and the second target privacy paragraph identifier;

Determine a privacy paragraph identifier statistic value of the first target privacy paragraph identifier and the second target privacy paragraph identifier;

The privacy paragraph identifier commonality score of the first target text set and the second target text set is determined based on the number of identical privacy paragraph identifiers of the first target privacy paragraph identifier and the second target privacy paragraph identifier, and the privacy paragraph identifier statistics of the first target privacy paragraph identifier and the second target privacy paragraph identifier.

In some embodiments, the first privacy paragraph text includes first statistical data of a target privacy paragraph included in the medical record text to be desensitized, and the second privacy paragraph text includes second statistical data of the target privacy paragraph included in the medical record text to be desensitized; wherein, based on the privacy paragraph topic commonality scores of the medical record text to be desensitized and the desensitized text of the patient's medical record, determining the privacy paragraph commonality scores of the medical record text to be desensitized and the desensitized text of the patient's medical record includes:

Based on the first statistical data and the second statistical data, determine a sum and a difference of the number of target private paragraphs in the medical record text to be desensitized and the desensitized medical record text;

Determine a commonality score of the number of private paragraphs in the medical record text to be desensitized and the desensitized text of the patient's medical record based on the sum and difference of the number of target private paragraphs in the medical record text to be desensitized and the desensitized text of the patient's medical record;

Based on the commonality score of the number of private paragraphs and the commonality score of the subject of private paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record, the commonality score of private paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record is determined.

In some embodiments, the medical record text to be desensitized includes a first medical record text to be desensitized and a second medical record text to be desensitized, the text commonality score between the first medical record text to be desensitized and the desensitized text of the patient medical record is a first text commonality score, and the text commonality score between the second medical record text to be desensitized and the desensitized text of the patient medical record is a second text commonality score; wherein, based on the text commonality score between the medical record text to be desensitized and the desensitized text of the patient medical record, determining the target desensitized text from the desensitized text of the patient medical record includes:

Respectively obtaining text semantic features of the first medical record text of the disease to be desensitized and the second medical record text of the disease to be desensitized;

Determining a second semantic commonality score between the first medical record text of the disease to be desensitized and the second medical record text of the disease to be desensitized based on text semantic features of the first medical record text of the disease to be desensitized and the second medical record text of the disease to be desensitized;

Based on the first text commonality score, the second text commonality score, and the second semantic commonality score, a target desensitized text of the first medical record text to be desensitized is determined from the desensitized text of the medical record.

In some embodiments, based on the first text commonality score, the second text commonality score, and the second semantic commonality score, determining a target desensitized text of the first medical record text to be desensitized from the desensitized medical record text includes:

Determine contribution information of the second medical record text of the disease to be desensitized to the first medical record text of the disease to be desensitized based on the second text commonality score and the second semantic commonality score;

Determine the correlation between the first to-be-desensitized patient medical record text and the desensitized patient medical record text based on the contribution information of the second to-be-desensitized patient medical record text to the first to-be-desensitized patient medical record text and the first text commonality score;

Based on the correlation between the first medical record text to be desensitized and the desensitized text of the patient's medical record, a target desensitized text of the first medical record text to be desensitized is determined from the desensitized text of the patient's medical record.

In some embodiments, performing privacy desensitization protection on the medical record text to be desensitized based on the target desensitized text includes:

Perform text matching processing on the target desensitized text and the medical record text to be desensitized to obtain a text matching result;

In combination with the text matching result, the medical record text to be desensitized is subjected to privacy desensitization protection through the target desensitized text to obtain a basic desensitized text;

The basic desensitized text is optimized by a text anonymization algorithm to perform privacy desensitization protection on the medical record text to be desensitized.

In some embodiments, before optimizing the basic desensitized text by a text anonymization algorithm to perform privacy desensitization protection on the medical record text to be desensitized, the method further includes:

Get the target anonymized text sample;

Performing privacy paragraph de-anonymization processing on the target anonymized text example to obtain a de-anonymized text example;

Loading the de-anonymized text example into a target machine learning algorithm to obtain anonymous predicted text;

determining a target debugging cost between the anonymous predicted text and the target anonymized text example;

The algorithm variables of the target machine learning algorithm are improved by using the target debugging cost to debug the target machine learning algorithm into the text anonymization algorithm.

In a second aspect, the present invention also provides an AI desensitization processing system, comprising a processor and a memory; the processor and the memory are communicatively connected, and the processor is used to read a computer program from the memory and execute it to implement the above method.

In a third aspect, the present invention further provides a computer-readable storage medium having a program stored thereon, which implements the above method when executed by a processor.

The patient information desensitization processing method and system combined with artificial intelligence provided by the embodiment of the present invention, when determining the target desensitized text for the medical record text to be desensitized, not only the global commonality between the target desensitized text and the medical record text to be desensitized is ensured through the text semantic features of the medical record text to be desensitized and the desensitized text of the patient's medical record; but also the commonality of the privacy paragraphs in the target desensitized text and the privacy paragraphs in the medical record text to be desensitized is ensured through the commonality score of the privacy paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record, that is, the similarity of the text information in the target desensitized text and the medical record text to be desensitized is ensured. In this way, the embodiment of the present invention can not only ensure the similarity of the text layout of the target desensitized text and the medical record text to be desensitized at the overall level, but also ensure the similarity of the privacy paragraph information in the target desensitized text and the medical record text to be desensitized, and improve the matching between the target desensitized text and the medical record text to be desensitized, so as to improve the data anonymity/desensitization quality and efficiency of the medical record text to be desensitized when the medical record text to be desensitized is anonymized/desensitized based on the target desensitized text.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

FIG1 is a flow chart of a method for desensitizing patient information combined with artificial intelligence provided in an embodiment of the present invention.

Implementation

Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

The method embodiments provided in the embodiments of the present invention can be executed in an AI desensitization processing system, a computer device or a similar computing device. Taking the operation on the AI desensitization processing system as an example, the AI desensitization processing system may include one or more processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory for storing data. Optionally, the above-mentioned AI desensitization processing system may also include a transmission device for communication functions. It can be understood by those of ordinary skill in the art that the above-mentioned structure is only illustrative and does not limit the structure of the above-mentioned AI desensitization processing system. For example, the AI desensitization processing system may also include more or fewer components than those shown above, or have a configuration different from that shown above.

The memory can be used to store computer programs, for example, software programs and modules of application software, such as a computer program corresponding to a method for desensitizing patient information combined with artificial intelligence in an embodiment of the present invention. The processor executes various functional applications and data processing by running the computer program stored in the memory, that is, implementing the above method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories can be connected to the AI desensitization processing system via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The transmission device is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of the AI desensitization processing system. In one example, the transmission device includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so that it can communicate with the Internet. In one example, the transmission device can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly.

Based on this, please refer to Figure 1, which is a flow chart of a patient information desensitization processing method combined with artificial intelligence provided by an embodiment of the present invention. The method is applied to an AI desensitization processing system and may further include STEP11-STEP17.

STEP 11, obtain the medical records to be desensitized and the desensitized medical records of the patients.

In an embodiment of the present invention, the medical record text to be desensitized can be a medical record text of a patient that needs to be desensitized for privacy protection. The medical record text to be desensitized can be an outpatient medical record text, an inpatient medical record text, or even a surgical medical record text. The medical record text to be desensitized records the privacy information related to the patient. If the medical record text is shared for research needs in the later stage, the privacy information needs to be protected. Therefore, the embodiment of the present invention can realize the privacy desensitization protection processing of the medical record text to be desensitized.

In addition, the desensitized medical record text may be a medical record text that has completed anonymization/desensitization processing of privacy information. The desensitized medical record text is obtained based on a K-anonymity algorithm or other anonymity algorithms.

Furthermore, there may be at least one medical record text to be desensitized, and there may be at least one desensitized medical record text.

STEP 12, obtain the parsing results of the private paragraphs in the medical records to be desensitized and the parsing results of the private paragraphs in the desensitized medical records of the patients.

In a possible design idea, the result of analyzing the private paragraphs in the medical records to be desensitized may include the target private paragraph mined in the medical records to be desensitized, the identification information of the target private paragraph, the distribution characteristics (position) of the target private paragraph in the medical records to be desensitized, and statistical data, and any other information related to the target private paragraph that can be mined from the medical records to be desensitized. The target private paragraph may be one or more.

STEP 13, based on the parsing results of the privacy paragraphs in the medical records to be desensitized and the parsing results of the privacy paragraphs in the desensitized text of the patient's medical records, determine the commonality scores of the privacy paragraphs in the medical records to be desensitized and the desensitized text of the patient's medical records.

In a possible design concept, the commonality score of the private paragraphs of the medical records to be desensitized and the desensitized medical records of the patients may refer to the commonality score of the target private paragraphs included in the medical records to be desensitized and the desensitized medical records of the patients, and the commonality score of the private paragraphs may include at least one of the commonality score of the number of target private paragraphs, the commonality score of themes, the commonality score of distribution characteristics, or the commonality score of privacy levels, etc. Among them, the commonality score can be understood as similarity or similarity.

STEP 14, determine the text semantic features of the medical record text to be desensitized and the text semantic features of the desensitized medical record text.

In a possible design idea, the target machine learning algorithm can be used to perform text vector mining on the medical records to be desensitized and the medical records of patients with desensitized texts, so as to obtain a text semantic feature that can characterize the overall situation of the medical records to be desensitized and a text semantic feature that can characterize the overall situation of the medical records with desensitized texts. The target machine learning algorithm can be a deep learning model DNN, residual network, etc.

STEP 15, based on the text semantic features of the medical records to be desensitized and the text semantic features of the desensitized text of the patient's medical records, determine the first semantic commonality score of the medical records to be desensitized and the desensitized text of the patient's medical records.

In a possible design idea, the semantic commonality score between the text semantic features of the medical records to be desensitized and the text semantic features of the desensitized medical records of the patients can be determined, and the semantic commonality score is used as the first semantic commonality score. The larger the first semantic commonality score, the greater the semantic feature matching (semantic feature correlation) between the medical records to be desensitized and the desensitized medical records of the patients.

Among them, the commonality score between the text semantic features of the medical record text to be desensitized and the text semantic features of the desensitized medical record text can be determined by, but is not limited to, cosine similarity.

For example, the text semantic feature of the medical record text to be desensitized is feature_u, and the text semantic feature of the patient medical record desensitized text is feature_v. Then the semantic commonality score between the text semantic features of the medical record text to be desensitized and the text semantic features of the patient medical record desensitized text can be determined by the dot product result of feature_u and feature_v. The larger the score obtained, the higher the macro commonality score between the medical record text to be desensitized and the patient medical record desensitized text. i is an integer not less than 1 and less than or equal to the number of medical record texts to be desensitized.

STEP 16, based on the privacy paragraph commonality score and the first semantic commonality score of the medical record text to be desensitized and the desensitized text of the patient's medical record, determine the text commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record.

Under one possible design idea, the privacy paragraph commonality score of the medical record text to be desensitized and the desensitized text of the patient's medical record can be integrated with the first semantic commonality score to serve as the text commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record. Alternatively, the privacy paragraph commonality score of the medical record text to be desensitized and the desensitized text of the patient's medical record can be multiplied with the first semantic commonality score to serve as the text commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record.

Based on the text commonality score determined by the above method, not only the commonality score relationship between the medical records to be desensitized and the desensitized text of the patient's medical records is introduced through the first semantic commonality score, but also the similarity of local text information in the medical records to be desensitized and the desensitized text of the patient's medical records is introduced through the privacy paragraph commonality score.

STEP 17, based on the text commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record, determine the target desensitized text from the desensitized text of the patient's medical record, so as to perform privacy desensitization protection on the medical record text to be desensitized based on the target desensitized text.

Under one possible design idea, the target desensitized text can be determined from the desensitized texts of each patient's medical record based on the text commonality scores of the medical record text to be desensitized and the desensitized texts of each patient's medical record. For example, the desensitized text of the patient's medical record corresponding to the maximum text commonality score can be determined in the text commonality score list as the target desensitized text of the medical record text to be desensitized.

For example, if there are two medical records to be desensitized, the privacy paragraph commonality score list List1 of the medical records to be desensitized and the desensitized texts of each patient's medical records, and the semantic commonality score list List2 of the medical records to be desensitized and the desensitized texts of each patient's medical records can be transformed from one dimension to two dimensions. After multiplying the privacy paragraph commonality score list List1 and the semantic commonality score list List2, the text commonality scores of the two medical records to be desensitized and the desensitized texts of each patient's medical records can be determined, and then the target desensitized texts can be determined for the two medical records to be desensitized, respectively.

In the embodiment of the present invention, when determining the target desensitized text for the medical record text to be desensitized, not only the global similarity between the target desensitized text and the medical record text to be desensitized is ensured through the text semantic features of the medical record text to be desensitized and the desensitized text of the patient's medical record; but also the similarity between the privacy paragraphs in the target desensitized text and the privacy paragraphs in the medical record text to be desensitized is ensured through the commonality score of the privacy paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record, that is, the similarity of the text information (such as the number of privacy paragraphs, the distribution of privacy paragraphs, etc.) in the target desensitized text and the medical record text to be desensitized is ensured. In this way, the embodiment of the present invention can not only ensure the similarity of the text layout of the target desensitized text and the medical record text to be desensitized at the overall level, but also ensure the similarity of the privacy paragraph information in the target desensitized text and the medical record text to be desensitized, significantly improving the matching between the target desensitized text and the medical record text to be desensitized, so as to improve the data anonymity/desensitization quality and efficiency of the medical record text to be desensitized when the medical record text to be desensitized is anonymized/desensitized based on the target desensitized text.

Under a possible design idea, after the target desensitized text of the medical record text to be desensitized is determined, the privacy desensitization protection of the medical record text to be desensitized can be completed in the following way.

The target desensitized text and the medical record text to be desensitized are subjected to text matching processing to obtain text matching results. Among them, the text matching processing can match the words and sentences of each private paragraph. For example, if the target desensitized text and the medical record text to be desensitized both have two private paragraphs with a corresponding relationship, through text matching processing, the names, home addresses, work units, daily living habits before and after illness and other words and sentences in the private paragraphs of the target desensitized text can be matched one by one with the names, home addresses, work units, daily living habits before and after illness and other words and sentences in the private paragraphs of the medical record text to be desensitized.

Combined with the text matching results, the target desensitized text can be used to perform privacy desensitization protection on the medical record text to be desensitized to obtain the basic desensitized text. The basic desensitized text is a text that has undergone preliminary anonymous desensitization processing. After obtaining the basic desensitized text, the basic desensitized text can be optimized (further privacy desensitization processing) through the debugged text anonymity algorithm to complete the privacy desensitization protection of the medical record text to be desensitized. In this way, through two rounds of desensitization processing, the smoothness of privacy anonymity can be achieved, avoiding excessive anonymity that causes the relevant text information to be too generalized and difficult to be used by later medical research institutes, that is, the embodiment of the present invention can maximize the usability of the desensitized medical record text while protecting user privacy.

Before optimizing the basic desensitized text through the debugged text anonymization algorithm to complete the privacy desensitization protection of the desensitized medical record text, the debugging of the text anonymization algorithm can be completed as follows: obtain a target anonymized text example; perform privacy paragraph de-anonymization on the target anonymized text example to obtain a de-anonymized text example; load the de-anonymized text example into the target machine learning algorithm to obtain anonymous predicted text; determine the target debugging cost between the anonymous predicted text and the target anonymized text example; improve the algorithm variables of the target machine learning algorithm through the target debugging cost to debug the target machine learning algorithm into a text anonymization algorithm.

Among them, the above text examples can be understood as training texts, and the debugging cost can be understood as the training loss of the algorithm.

In an embodiment of the present invention, the privacy paragraph parsing results in the medical record text to be desensitized may include a first privacy paragraph text mined from the medical record text to be desensitized, and the privacy paragraph parsing results in the desensitized medical record text may include a second privacy paragraph text mined from the desensitized medical record text.

Among them, the first privacy paragraph text may include statistical data of the target privacy paragraph in the patient's medical record text to be desensitized, and the second privacy paragraph text may include statistical data of the target privacy paragraph in the patient's medical record desensitized text.

Under one possible design idea, the target privacy paragraph mining network can be used to perform medical record text mining on the medical record text to be desensitized or the patient medical record desensitized text to determine the first privacy paragraph text or the second privacy paragraph text from the medical record text to be desensitized or the patient medical record desensitized text.

Based on this, the determination of the commonality score of the above-mentioned privacy paragraphs includes the following steps.

STEP 21, determine first statistical data of the target private paragraph in the medical record text to be desensitized based on the first private paragraph text.

STEP 22, determine the second statistical data of the target private paragraph in the desensitized text of the patient's medical record based on the second private paragraph text.

STEP23, based on the first statistical data and the second statistical data, determine the sum and difference of the number of target privacy paragraphs in the medical record text to be desensitized and the desensitized medical record text of the patient.

Here, statistical data can be understood as quantity or number.

STEP24, based on the sum and difference of the number of target private paragraphs in the medical records to be desensitized and the desensitized text of the patient's medical records, determine the commonality score of the number of private paragraphs in the medical records to be desensitized and the desensitized text of the patient's medical records.

In a possible design idea, the commonality score of the number of private paragraphs between the medical record text to be desensitized and the desensitized medical record text of the patient can be determined based on the weighted average result of the number summation result and the number difference result. Among them, the specific parameters of the weighted average can be adjusted by those skilled in the art according to actual needs.

STEP 25, based on the commonality score of the number of private paragraphs in the medical records to be desensitized and the desensitized text of the patient's medical records, determine the commonality score of private paragraphs in the medical records to be desensitized and the desensitized text of the patient's medical records.

In a possible design idea, the commonality score of the number of private paragraphs between the medical record text to be desensitized and the patient's medical record desensitized text can be used as the text commonality score between the medical record text to be desensitized and the patient's medical record desensitized text. For example, the commonality score list of the number of private paragraphs of the medical record text to be desensitized and the patient's medical record desensitized text is used as the commonality score list of private paragraphs between the medical record text to be desensitized and the patient's medical record desensitized text.

It can be seen that through the embodiments of the present invention, not only can the global similarity between the target desensitized text and the medical record text to be desensitized be guaranteed through the semantic commonality scoring of the medical record text, but also the number of target privacy paragraphs in the target desensitized text and the medical record text to be desensitized can be guaranteed to be consistent, so that when the medical record text to be desensitized is anonymized/desensitized through the target desensitized text, the target privacy paragraphs in the medical record text to be desensitized all have corresponding privacy paragraphs in the target desensitized text, thereby ensuring the data anonymity/desensitization quality and efficiency of each target privacy paragraph in each medical record text to be desensitized.

Under other design ideas, the determination of the above-mentioned privacy paragraph analysis results may include the following content.

STEP31, perform privacy paragraph mining processing on the medical record text to be desensitized and the desensitized text of the patient's medical record, respectively, to determine the first privacy paragraph text included in the medical record text to be desensitized and the second privacy paragraph text included in the desensitized text of the patient's medical record.

The first privacy paragraph text may include the target privacy paragraphs in the patient medical record text to be desensitized and the distribution characteristics of each target privacy paragraph in the patient medical record text to be desensitized. The second privacy paragraph text may include the target privacy paragraphs in the patient medical record desensitized text and the distribution characteristics of each target privacy paragraph in the patient medical record desensitized text.

In the embodiment of the present invention, there may be multiple target privacy paragraphs in a patient medical record text to be desensitized or a patient medical record desensitized text, and each target privacy paragraph may belong to a different privacy paragraph identifier.

Under one possible design idea, the debugged privacy paragraph mining machine learning algorithm can be used to mine the desensitized medical record text, so as to mine each target privacy paragraph and the distribution characteristics of each target privacy paragraph from the desensitized medical record text or the desensitized medical record text.

STEP 32, based on the first privacy paragraph text and the second privacy paragraph text, determine the privacy paragraph identifier to which each text unit in the medical record text to be desensitized belongs and the privacy paragraph identifier to which each text unit in the desensitized medical record text belongs.

In some examples, after the distribution characteristics of each target private paragraph are mined in the medical record text to be desensitized or the desensitized text of the patient's medical record, each text unit in the medical record text to be desensitized or the desensitized text of the patient's medical record is equivalent to matching its respective identification field.

For example, target privacy paragraphs with three different privacy paragraph markers, Q, W, and E, are mined from the medical record text to be desensitized. Then a certain text unit in the medical record text to be desensitized either belongs to the Q privacy paragraph marker, the W privacy paragraph marker, or the E privacy paragraph marker, and of course may also belong to other markers.

STEP 33, decompose the medical record text to be desensitized into X first text sets, and desensitize the medical record text into X second text sets, the X first text sets correspond to the X second text sets one by one, and X is an integer not less than 1.

In a possible design idea, the desensitized patient medical record text or the patient medical record desensitized text can be decomposed into X text sets according to certain decomposition rules. The embodiment of the present invention does not limit the above decomposition method and the number of patient medical record text sets obtained by decomposition, but it is necessary to ensure that the decomposition method of the patient medical record text to be desensitized is consistent with the desensitized patient medical record text, so as to ensure that each patient medical record text set in the patient medical record text to be desensitized corresponds to the patient medical record text set in the patient medical record desensitized text.

STEP34, based on the privacy paragraph identifiers of each text unit in the medical record text to be desensitized, determine the first privacy paragraph identifiers of each text unit in each first text set as the privacy paragraph parsing result in the medical record text to be desensitized.

STEP35, based on the privacy paragraph identifiers of each text unit in the desensitized text of the patient medical record, determine the second privacy paragraph identifiers of each text unit in each second text set as the privacy paragraph parsing result in the desensitized text of the patient medical record.

Under one possible design idea, after obtaining the first privacy paragraph identifier belonging to each text unit in each first text set and the second privacy paragraph identifier belonging to each text unit in each second text set, the privacy paragraph parsing results in the medical record text to be desensitized and the desensitized text of the patient's medical record can be determined based on the first privacy paragraph identifier in each first text set and the second privacy paragraph identifier in each second text set.

In one possible design idea, the commonality score of the text unit status identifiers between each first text set and the corresponding second text set can be counted based on the first privacy paragraph identifiers in each first text set and the second privacy paragraph identifiers in each second text set, and then the commonality score of the privacy paragraphs between the medical records to be desensitized and the desensitized text of the patient's medical records can be determined based on the commonality score of the text unit status identifiers corresponding to each first text set. For example, the commonality score of the text unit status identifiers of each first text set and its corresponding second text set can be averaged to determine the commonality score of the privacy paragraphs between the medical records to be desensitized and the desensitized text of the patient's medical records.

Among them, the commonality score of the text unit status identification (i.e., the privacy paragraph identification field corresponding to the text unit) of the first text set and its corresponding second text set can be determined in the following way: determining the number of text units in the first text set and the second text set with the same status; determining a ratio value by the number of text units in the first text set and the second text set with the same status and the statistical value of the text units of the first text set (or the second text set), and using the ratio value as the commonality score of the text unit status identification of the first text set and its corresponding second text set.

Under some other possible design ideas, the determination of the commonality score of the above-mentioned privacy paragraphs may include the following content.

STEP 41, based on the first privacy paragraph identifiers belonging to each text unit in each first text set and the second privacy paragraph identifiers belonging to each text unit in each second text set, determine the privacy paragraph identifier commonality score between each first text set and the corresponding second text set.

In a possible design concept, if the first privacy paragraph identifier to which each text unit in each first text set belongs is known, the number of privacy paragraph identifiers included in each first text set can be counted. Similarly, if the second privacy paragraph identifier to which each text unit in each second text set belongs is known, the number of privacy paragraph identifiers included in each second text set can be counted.

Generally, given that the disassembly method of the u-th patient medical record text to be desensitized is the same as the disassembly method of the v-th patient medical record text to be desensitized, there is a one-to-one correspondence between the first text set of the u-th patient medical record text to be desensitized and the second text set of the v-th patient medical record text to be desensitized, so the p-th first text set of the u-th patient medical record text to be desensitized corresponds to the p-th second text set of the v-th patient medical record desensitized text.

In some cases, the privacy paragraph identifier commonality score of the first target text set and the second target text set can be determined in the following manner: determining the number of identical privacy paragraph identifiers of the first target privacy paragraph identifier and the second target privacy paragraph identifier; determining the privacy paragraph identifier statistics of the first target privacy paragraph identifier and the second target privacy paragraph identifier; determining the privacy paragraph identifier commonality score of the first target text set and the second target text set based on the number of identical privacy paragraph identifiers of the first target privacy paragraph identifier and the second target privacy paragraph identifier and the privacy paragraph identifier statistics of the first target privacy paragraph identifier and the second target privacy paragraph identifier.

In a possible design idea, the commonality score of the privacy paragraph identifiers of each first text set and the corresponding second text set can be determined based on the method for determining the commonality score of the privacy paragraph identifiers of the first target text set and the second target text set.

STEP 42, based on the privacy paragraph identification commonality scores between each first text set and the corresponding second text set, determine the privacy paragraph topic commonality scores of the patient's medical record text to be desensitized and the patient's medical record desensitized text.

Under one possible design idea, a series of operations can be performed on the commonality scores of the privacy paragraph identification of each first text set and the corresponding second text set to determine the commonality scores of the privacy paragraph topics of the medical record text to be desensitized and the desensitized medical record text of the patient.

STEP43, based on the commonality scores of the topics of the private paragraphs in the medical records to be desensitized and the desensitized texts of the patients' medical records, determine the commonality scores of the private paragraphs in the medical records to be desensitized and the desensitized texts of the patients' medical records.

In a possible design idea, the privacy paragraph theme commonality score of the medical record text to be desensitized and the patient medical record desensitized text can be used as the privacy paragraph commonality score of the medical record text to be desensitized and the patient medical record desensitized text. For example, the privacy paragraph theme commonality score of the medical record text to be desensitized and the patient medical record desensitized text is used as the privacy paragraph commonality score list between the medical record text to be desensitized and the patient medical record desensitized text.

Under one possible design idea, the first private paragraph text mined from the medical record text to be desensitized may include first statistical data of the target private paragraph included in the medical record text to be desensitized, and the second private paragraph text mined and processed from the desensitized medical record text may include second statistical data of the target private paragraph included in the medical record text to be desensitized.

Under other design ideas, the determination of the commonality score of the above-mentioned privacy paragraphs may include the following content.

STEP 51, based on the first privacy paragraph identifiers belonging to each text unit in each first text set and the second privacy paragraph identifiers belonging to each text unit in each second text set, determine the privacy paragraph identifier commonality score between each first text set and the corresponding second text set.

STEP52, based on the privacy paragraph identification commonality scores between each first text set and the corresponding second text set, determine the privacy paragraph topic commonality scores of the patient's medical record text to be desensitized and the patient's medical record desensitized text.

STEP53, based on the first statistical data and the second statistical data, determine the sum and difference of the number of target privacy paragraphs in the medical record text to be desensitized and the desensitized medical record text of the patient.

STEP54, based on the sum and difference of the number of target private paragraphs in the medical records to be desensitized and the desensitized text of the patient's medical records, determine the commonality score of the number of private paragraphs in the medical records to be desensitized and the desensitized text of the patient's medical records.

STEP55, based on the commonality score of the number of private paragraphs between the medical records to be desensitized and the desensitized text of the patient's medical records and the commonality score of the topics of private paragraphs, determine the commonality score of private paragraphs between the medical records to be desensitized and the desensitized text of the patient's medical records.

In one possible design idea, the privacy paragraph theme commonality score and the privacy paragraph number commonality score of the medical record text to be desensitized and the patient medical record desensitized text can be multiplied or summed to determine the privacy paragraph commonality score of the medical record text to be desensitized and the patient medical record desensitized text. For example, the privacy paragraph theme commonality score list of the medical record text to be desensitized and the patient medical record desensitized text and the privacy paragraph number commonality score list can be multiplied or summed to obtain the privacy paragraph commonality score list between the medical record text to be desensitized and the patient medical record desensitized text.

The embodiments of the present invention can not only ensure the global similarity between the target desensitized text and the medical record text to be desensitized through the semantic commonality scoring of the medical record text; it can also ensure that the number of target privacy paragraphs in the target desensitized text and the medical record text to be desensitized is consistent, so that when the medical record text to be desensitized is anonymized/desensitized by the target desensitized text, the data anonymity/desensitization quality and efficiency of the target privacy paragraphs can be guaranteed; it can also ensure that the target desensitized text has privacy paragraph themes similar to the first text set in the distribution area corresponding to the first text set through the privacy paragraph theme commonality scoring.

Through the embodiments of the present invention, not only can the global similarity of the target desensitized text and the medical record text to be desensitized be ensured through the semantic commonality scoring of the medical record text, but also the consistency of the target number of private paragraphs in the target desensitized text and the medical record text to be desensitized can be ensured; the consistency of the distribution of private paragraphs in the target desensitized text and the target private paragraphs can also be ensured, that is, the consistency of the private paragraph themes in the target desensitized text and the medical record text to be desensitized in the corresponding distribution areas is ensured.

Under one possible design idea, the medical record text to be desensitized includes a first medical record text to be desensitized and a second medical record text to be desensitized, wherein the text commonality score between the first medical record text to be desensitized and the desensitized medical record text can be a first text commonality score score1v, and the text commonality score between the second medical record text to be desensitized and the desensitized medical record text can be a second text commonality score score2v, wherein v is an integer greater than or equal to 1 and less than or equal to the number of desensitized medical record texts.

Furthermore, the determination of the above-mentioned target desensitized text may include the following contents.

STEP 61, respectively obtain text semantic features of the medical record text of the first disease to be desensitized and the medical record text of the second disease to be desensitized.

Under one possible design idea, some target machine learning algorithms can be used to perform text vector mining on the medical records of the first patient to be desensitized and the medical records of the second patient to be desensitized to obtain a text semantic feature that can characterize the overall situation of the medical records of the first patient to be desensitized and a text semantic feature that can characterize the overall situation of the medical records of the second patient to be desensitized.

Among them, the target machine learning algorithm can be a deep learning model DNN, a residual network, etc.

STEP62, based on the text semantic features of the first medical record text of the disease to be desensitized and the second medical record text of the disease to be desensitized, determine the second semantic commonality score between the first medical record text of the disease to be desensitized and the second medical record text of the disease to be desensitized.

In a possible design concept, a commonality score between the text semantic features of the first medical record text to be desensitized and the text semantic features of the second medical record text to be desensitized can be determined as the second semantic commonality score.

Among them, the commonality score between the text semantic features of the first medical record text to be desensitized and the text semantic features of the second medical record text to be desensitized can be calculated by cosine similarity.

Under one possible design idea, after obtaining the semantic commonality scores of the first medical record text to be desensitized and the second medical record text to be desensitized, the target desensitized text of the first medical record text to be desensitized can be determined from the desensitized texts of the medical records based on the first text commonality score, the second text commonality score and the second semantic commonality score.

Under one possible design idea, the method shown in STEP63-STEP65 can be used to determine the target desensitized text of the first medical record text to be desensitized from the desensitized text of the medical record based on the first text commonality score, the second text commonality score and the second semantic commonality score.

STEP 63: Determine contribution information of the second medical record text to be desensitized to the first medical record text to be desensitized based on the second text commonality score and the second semantic commonality score.

Under one possible design idea, the second semantic commonality score between the second medical record text to be desensitized and the first medical record text to be desensitized can be multiplied or added with the second text commonality score between the second medical record text to be desensitized and the desensitized medical record text to determine the contribution information of the second medical record text to be desensitized relative to the first medical record text to be desensitized.

STEP64, based on the contribution information of the second medical record text to be desensitized to the first medical record text to be desensitized and the commonality score of the first text, determine the correlation between the first medical record text to be desensitized and the desensitized text of the patient's medical record.

Under one possible design idea, the contribution information of the second patient medical record text to the first patient medical record text to be desensitized can be summed with the first text commonality score of the first patient medical record text to be desensitized and the patient's desensitized medical record text to determine the correlation between the first patient medical record text to be desensitized and the patient's desensitized medical record text.

STEP65, based on the correlation between the first patient medical record text to be desensitized and the patient medical record desensitized text, determine the target desensitized text of the first patient medical record text to be desensitized from the patient medical record desensitized text.

Under one possible design idea, the correlation between the first patient's medical record text to be desensitized and multiple patient's medical record desensitized texts can be determined through the method of the embodiment of the present invention, and then the patient's medical record desensitized text corresponding to the maximum correlation is determined from the multiple correlations as the target desensitized text of the first patient's medical record text to be desensitized.

Among them, the correlation between the first medical record text to be desensitized and the desensitized medical record text of the patient not only introduces the medical record text features of the first medical record text to be desensitized and the desensitized medical record text of the patient, but also introduces the influence of the commonality degree of the second medical record text to be desensitized on the first medical record text to be desensitized.

Under one possible design idea, after the patient medical record text correlation between the patient medical record text to be desensitized and the desensitized texts of each patient medical record is determined by the above method, the patient medical record desensitized text corresponding to the maximum patient medical record text correlation can be used as the target desensitized text of the patient medical record text to be desensitized.

Another patient information desensitization processing method combined with artificial intelligence provided by an embodiment of the present invention may include the following content.

STEP 71, obtain the medical record text of the first patient to be desensitized, the medical record text of the second patient to be desensitized, and the desensitized text of the patient's medical record.

STEP 72, obtain the first text commonality score between the first patient medical record text to be desensitized and the patient medical record desensitized text.

The first text commonality score can be a semantic commonality score between the first medical record text to be desensitized and the desensitized text of the patient's medical record, or it can be other commonality scores related to the content of the patient's medical record text determined based on the semantic commonality score between the first medical record text to be desensitized and the desensitized text of the patient's medical record, for example, it can be a commonality score determined based on the privacy paragraph commonality score and the semantic commonality score between the first medical record text to be desensitized and the desensitized text of the patient's medical record.

The semantic commonality score between the first medical record text to be desensitized and the desensitized text of the patient's medical record can be determined in the following manner: determine the text semantic features of the first medical record text to be desensitized and the text semantic features of the desensitized text of the patient's medical record; determine the semantic commonality score between the first medical record text to be desensitized and the desensitized text of the patient's medical record based on the text semantic features of the first medical record text to be desensitized and the text semantic features of the desensitized text of the patient's medical record.

The privacy paragraph commonality score between the first patient medical record text to be desensitized and the patient medical record desensitized text can be determined in the following manner: mining the first privacy paragraph text from the first patient medical record text to be desensitized, and mining the second privacy paragraph text from the patient medical record desensitized text; determining the first statistical data of the target privacy paragraph in the first patient medical record text to be desensitized based on the first privacy paragraph text; determining the second statistical data of the target privacy paragraph in the patient medical record desensitized text based on the second privacy paragraph text; determining the sum and difference results of the number of target privacy paragraphs in the first patient medical record text to be desensitized and the patient medical record desensitized text based on the first statistical data and the second statistical data; determining the commonality score of the number of privacy paragraphs between the first patient medical record text to be desensitized and the patient medical record desensitized text based on the sum and difference results of the number of target privacy paragraphs in the first patient medical record text to be desensitized and the patient medical record desensitized text; determining the privacy paragraph commonality score between the first patient medical record text to be desensitized and the patient medical record desensitized text based on the commonality score of the number of privacy paragraphs between the first patient medical record text to be desensitized and the patient medical record desensitized text.

The privacy paragraph commonality score between the first medical record text to be desensitized and the patient medical record desensitized text can also be determined in the following manner: based on the first privacy paragraph identifier belonging to each text unit in each first text set and the second privacy paragraph identifier belonging to each text unit in each second text set, determine the privacy paragraph identifier commonality score between each first text set and the corresponding second text set; based on the privacy paragraph identifier commonality score between each first text set and the corresponding second text set, determine the privacy paragraph topic commonality score between the first medical record text to be desensitized and the patient medical record desensitized text; based on the privacy paragraph topic commonality score between the first medical record text to be desensitized and the patient medical record desensitized text, determine the privacy paragraph commonality score between the first medical record text to be desensitized and the patient medical record desensitized text.

The privacy paragraph commonality score between the first medical record text to be desensitized and the patient medical record desensitized text can also be determined in the following manner: based on the first privacy paragraph identifier belonging to each text unit in each first text set and the second privacy paragraph identifier belonging to each text unit in each second text set, determine the privacy paragraph identifier commonality score between each first text set and the corresponding second text set; based on the privacy paragraph identifier commonality score between each first text set and the corresponding second text set, determine the privacy paragraph topic commonality score between the first medical record text to be desensitized and the patient medical record desensitized text; based on the first statistical data and the second 2. Statistical data, determine the sum and difference of the number of target private paragraphs in the first medical record to be desensitized and the desensitized text of the patient's medical record; determine the commonality score of the number of private paragraphs in the first medical record to be desensitized and the desensitized text of the patient's medical record based on the sum and difference of the number of target private paragraphs in the first medical record to be desensitized and the desensitized text of the patient's medical record; determine the commonality score of private paragraphs in the first medical record to be desensitized and the desensitized text of the patient's medical record based on the commonality score of the number of private paragraphs and the commonality score of the theme of private paragraphs between the first medical record to be desensitized and the desensitized text of the patient's medical record.

STEP73, obtain the second text commonality score between the second patient's medical record text to be desensitized and the patient's medical record desensitization text.

The second text commonality score can be a semantic commonality score between the second medical record text to be desensitized and the desensitized text of the patient's medical record, or it can be other commonality scores related to the content of the patient's medical record text determined based on the semantic commonality score between the second medical record text to be desensitized and the desensitized text of the patient's medical record, for example, it can be a commonality score determined based on the privacy paragraph commonality score and the semantic commonality score between the second medical record text to be desensitized and the desensitized text of the patient's medical record.

The method for determining the privacy paragraph commonality score and semantic commonality score between the second medical record text to be desensitized and the patient's medical record desensitized text is similar to the method for determining the privacy paragraph commonality score and semantic commonality score between the first medical record text to be desensitized and the patient's medical record desensitized text.

STEP 74, respectively obtain the semantic features of the first medical record text to be desensitized and the second text to be desensitized.

STEP75, based on the semantic features of the first medical record text to be desensitized and the second medical record text to be desensitized, determine the second semantic commonality score between the first medical record text to be desensitized and the second medical record text to be desensitized.

STEP76, based on the first text commonality score, the second text commonality score and the second semantic commonality score, determine the target desensitized text of the first medical record text to be desensitized from the desensitized text of the medical record, so as to perform privacy desensitization protection on the first medical record text to be desensitized based on the target desensitized text.

The embodiment of the present invention not only ensures the global similarity between the target desensitized text and the medical record text to be desensitized through the text semantic features of the medical record text to be desensitized and the desensitized text of the patient's medical record; it also ensures that the privacy paragraphs in the target desensitized text are similar to the privacy paragraphs in the medical record text to be desensitized through the commonality score of the privacy paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record, that is, the number of privacy paragraphs, privacy paragraph distribution, privacy paragraph themes, etc. in the target desensitized text and the medical record text to be desensitized are similar. In this way, the embodiment of the present invention can not only ensure the similarity of the text layout of the target desensitized text and the medical record text to be desensitized at the overall level, but also ensure the similarity of the privacy paragraph information in the target desensitized text and the medical record text to be desensitized, which significantly improves the matching between the target desensitized text and the medical record text to be desensitized.

Under some independent design ideas, after determining the target desensitized text from the desensitized text of the patient medical records based on the text commonality score between the medical record text to be desensitized and the desensitized text of the patient medical records, so as to perform privacy desensitization protection on the medical record text to be desensitized based on the target desensitized text, the method also includes STEP18.

STEP18, in response to the sharing request of the desensitized medical record text for the medical record text to be desensitized, perform risk detection on the target digital medical server, and share the desensitized medical record text with the target digital medical server on the basis that the target digital medical server passes the risk detection.

It can be seen that before sharing the desensitized medical records, targeted risk detection will be carried out on the target digital medical server to ensure the security of sharing the desensitized medical records.

Under some independent design ideas, the above-mentioned risk detection of the target digital medical server includes the following contents.

STEP 181, obtain the data risk detection log of the target digital medical server, load the data risk detection log into the risk assessment strategy, and extract the risk detection description vector of the data risk detection log in the risk assessment strategy.

Among them, the risk assessment strategy can be a decision tree model for risk identification.

STEP 182, input the risk detection description vector into the risk decision module and the state deduction module in the risk assessment strategy respectively. The risk decision module includes first execution configuration data biased towards risk category decision, and the state deduction module includes second execution configuration data biased towards session state deduction.

Among them, the execution configuration data can guide related modules to perform feature processing.

STEP 183: In the risk decision module, the risk detection description vector is processed by using the first execution configuration data to obtain a category decision vector relationship network.

STEP 184: In the state deduction module, the risk detection description vector is processed by the second execution configuration data to obtain a state deduction vector relationship network.

STEP185, determine the risk detection index of the target digital medical server according to the category decision vector relationship network and the state deduction vector relationship network; on the basis that the risk detection index does not exceed the set detection index, determine that the target digital medical server has passed the risk detection; otherwise, determine that the target digital medical server has not passed the risk detection.

In the embodiment of the present invention, by performing feature operations (such as weighted processing) on the category decision vector relationship network and the state deduction vector relationship network, the risk detection index of the target digital medical server can be accurately calculated from the two levels of risk category and risk state. Thus, accurate risk detection quantitative judgment can be performed based on the risk detection index.

Furthermore, a computer-readable storage medium is provided, on which a program is stored, and when the program is executed by a processor, the above method is implemented.

In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method can also be implemented in other ways. The apparatus and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the apparatus, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and/or flowchart, and the combination of boxes in the block diagram and/or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code. It should be noted that in this article, the term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims (9)

1. A patient information desensitization processing method combined with artificial intelligence, characterized in that it is applied to an AI desensitization processing system, and the method includes: Obtain the medical records of the patients to be desensitized and the desensitized medical records of the patients; Obtain the parsing results of the private paragraphs in the medical record text to be desensitized and the parsing results of the private paragraphs in the desensitized medical record text; Determine the privacy paragraph commonality score of the medical record text to be desensitized and the desensitized medical record text based on the privacy paragraph parsing results in the medical record text to be desensitized and the privacy paragraph parsing results in the desensitized medical record text of the patient; Determine the text semantic features of the medical record text to be desensitized and the text semantic features of the desensitized medical record text; Determine a first semantic commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record based on the text semantic features of the medical record text to be desensitized and the text semantic features of the desensitized text of the patient's medical record; Determine a text commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record based on the privacy paragraph commonality score of the medical record text to be desensitized and the desensitized text of the patient's medical record and the first semantic commonality score; Based on the text commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record, determine the target desensitized text from the desensitized text of the patient's medical record, so as to perform privacy desensitization protection on the medical record text to be desensitized based on the target desensitized text; The privacy paragraph parsing result in the medical record text to be desensitized includes a first privacy paragraph text mined from the medical record text to be desensitized, and the privacy paragraph parsing result in the medical record desensitized text includes a second privacy paragraph text mined from the medical record desensitized text; based on the privacy paragraph parsing result in the medical record text to be desensitized and the privacy paragraph parsing result in the medical record desensitized text, the privacy paragraph commonality score of the medical record text to be desensitized and the medical record desensitized text is determined, including: Determine first statistical data of a target private paragraph in the medical record text to be desensitized based on the first private paragraph text; Determine second statistical data of the target private paragraph in the desensitized text of the patient's medical record based on the second private paragraph text; Based on the first statistical data and the second statistical data, determine a sum and a difference of the number of target private paragraphs in the medical record text to be desensitized and the desensitized medical record text; Determine a commonality score of the number of private paragraphs in the medical record text to be desensitized and the desensitized text of the patient's medical record based on the sum and difference of the number of target private paragraphs in the medical record text to be desensitized and the desensitized text of the patient's medical record; Based on the commonality score of the number of private paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record, the commonality score of the private paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record is determined. 2. The method according to claim 1, characterized in that obtaining the parsing results of the private paragraphs in the medical record text to be desensitized and the parsing results of the private paragraphs in the desensitized medical record text of the patient comprises: Performing privacy paragraph mining processing on the medical record text to be desensitized and the desensitized text of the patient's medical record, respectively, to determine a first privacy paragraph text included in the medical record text to be desensitized and a second privacy paragraph text included in the desensitized text of the patient's medical record; Based on the first privacy paragraph text and the second privacy paragraph text, determine the privacy paragraph identifier to which each text unit in the medical record text to be desensitized belongs and the privacy paragraph identifier to which each text unit in the medical record desensitized text belongs; Decomposing the medical record text to be desensitized into X first text sets, and decomposing the medical record desensitized text into X second text sets, wherein the X first text sets correspond to the X second text sets one by one, and X is an integer not less than 1; Determine the first privacy paragraph identifier of each text unit in each first text set based on the privacy paragraph identifier of each text unit in the medical record text to be desensitized, as the privacy paragraph parsing result in the medical record text to be desensitized; Based on the privacy paragraph identifiers of each text unit in the desensitized text of the patient's medical record, the second privacy paragraph identifiers of each text unit in each second text set are determined as the privacy paragraph parsing results in the desensitized text of the patient's medical record. 3. The method according to claim 2, characterized in that, based on the parsing results of the private paragraphs in the medical record text to be desensitized and the parsing results of the private paragraphs in the desensitized text of the patient's medical record, determining the commonality score of the private paragraphs of the medical record text to be desensitized and the desensitized text of the patient's medical record, comprises: Determine a privacy paragraph identifier commonality score between each first text set and the corresponding second text set based on a first privacy paragraph identifier to which each text unit belongs in each first text set and a second privacy paragraph identifier to which each text unit belongs in each second text set; Determine the privacy paragraph topic commonality score between the patient medical record text to be desensitized and the patient medical record desensitized text based on the privacy paragraph identification commonality score between each first text set and the corresponding second text set; Based on the commonality scores of the private paragraph topics of the medical record text to be desensitized and the desensitized text of the patient's medical record, the commonality scores of the private paragraphs of the medical record text to be desensitized and the desensitized text of the patient's medical record are determined. 4. The method as claimed in claim 3, characterized in that the first privacy paragraph identifier includes a first target privacy paragraph identifier, the second privacy paragraph identifier includes a second target privacy paragraph identifier, the first text set includes a first target text set, the second text set includes a second target text set, the first target text set corresponds to the second target text set, the text unit in the first target text set belongs to the first target privacy paragraph identifier, and the text unit in the second target text set belongs to the second target privacy paragraph identifier; wherein determining the privacy paragraph identifier commonality score between each first text set and the corresponding second text set comprises: Determine the number of identical privacy paragraph identifiers between the first target privacy paragraph identifier and the second target privacy paragraph identifier; Determine a privacy paragraph identifier statistic value of the first target privacy paragraph identifier and the second target privacy paragraph identifier; The privacy paragraph identifier commonality score of the first target text set and the second target text set is determined based on the number of identical privacy paragraph identifiers of the first target privacy paragraph identifier and the second target privacy paragraph identifier, and the privacy paragraph identifier statistics of the first target privacy paragraph identifier and the second target privacy paragraph identifier. 5. The method according to claim 3, characterized in that the first private paragraph text includes first statistical data of the target private paragraph included in the medical record text to be desensitized, and the second private paragraph text includes second statistical data of the target private paragraph included in the medical record text to be desensitized; wherein, based on the private paragraph topic commonality score of the medical record text to be desensitized and the desensitized text of the patient's medical record, determining the private paragraph commonality score of the medical record text to be desensitized and the desensitized text of the patient's medical record comprises: Based on the first statistical data and the second statistical data, determine a sum and a difference of the number of target private paragraphs in the medical record text to be desensitized and the desensitized medical record text; Determine a commonality score of the number of private paragraphs in the medical record text to be desensitized and the desensitized text of the patient's medical record based on the sum and difference of the number of target private paragraphs in the medical record text to be desensitized and the desensitized text of the patient's medical record; Based on the commonality score of the number of private paragraphs and the commonality score of the subject of private paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record, the commonality score of private paragraphs between the medical record text to be desensitized and the desensitized text of the patient's medical record is determined. 6. The method according to claim 1, characterized in that the medical record text to be desensitized includes a first medical record text to be desensitized and a second medical record text to be desensitized, the text commonality score between the first medical record text to be desensitized and the desensitized text of the patient's medical record is a first text commonality score, and the text commonality score between the second medical record text to be desensitized and the desensitized text of the patient's medical record is a second text commonality score; wherein, based on the text commonality score between the medical record text to be desensitized and the desensitized text of the patient's medical record, determining the target desensitized text from the desensitized text of the patient's medical record comprises: Respectively obtaining text semantic features of the first medical record text of the disease to be desensitized and the second medical record text of the disease to be desensitized; Determining a second semantic commonality score between the first medical record text of the disease to be desensitized and the second medical record text of the disease to be desensitized based on text semantic features of the first medical record text of the disease to be desensitized and the second medical record text of the disease to be desensitized; Based on the first text commonality score, the second text commonality score, and the second semantic commonality score, a target desensitized text of the first medical record text to be desensitized is determined from the desensitized text of the medical record. 7. The method according to claim 6, characterized in that the step of determining the target desensitized text of the first medical record text to be desensitized from the desensitized medical record text based on the first text commonality score, the second text commonality score, and the second semantic commonality score comprises: Determine contribution information of the second medical record text of the disease to be desensitized to the first medical record text of the disease to be desensitized based on the second text commonality score and the second semantic commonality score; Determine the correlation between the first to-be-desensitized patient medical record text and the desensitized patient medical record text based on the contribution information of the second to-be-desensitized patient medical record text to the first to-be-desensitized patient medical record text and the first text commonality score; Based on the correlation between the first to-be-desensitized medical record text and the desensitized medical record text of the patient, determining a target desensitized medical record text of the first to-be-desensitized medical record text from the desensitized medical record text of the patient; Wherein, performing privacy desensitization protection on the medical record text to be desensitized based on the target desensitized text includes: Perform text matching processing on the target desensitized text and the medical record text to be desensitized to obtain a text matching result; In combination with the text matching result, the medical record text to be desensitized is subjected to privacy desensitization protection through the target desensitized text to obtain a basic desensitized text; The basic desensitized text is optimized by a text anonymization algorithm to perform privacy desensitization protection on the medical record text to be desensitized; Wherein, before optimizing the basic desensitized text by a text anonymization algorithm to perform privacy desensitization protection on the medical record text to be desensitized, the method further includes: Get the target anonymized text sample; Performing privacy paragraph de-anonymization processing on the target anonymized text example to obtain a de-anonymized text example; Loading the de-anonymized text example into a target machine learning algorithm to obtain anonymous predicted text; determining a target debugging cost between the anonymous predicted text and the target anonymized text example; The algorithm variables of the target machine learning algorithm are improved by using the target debugging cost to debug the target machine learning algorithm into the text anonymization algorithm. 8. An AI desensitization processing system, characterized in that it includes a processor and a memory; the processor and the memory are communicatively connected, and the processor is used to read a computer program from the memory and execute it to implement the method described in any one of claims 1-7. 9. A computer-readable storage medium, characterized in that a program is stored thereon, and when the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.